Opacus v1.2

We're glad to present Opacus v1.2, which contains some major updates to per sample gradient computation mechanisms and includes all the good stuff from the recent PyTorch releases.

Highlights

Functorch - per sample gradients for all

With the recent release of functorch it's now easy to compute per sample gradients for any module, without any limitations we've had to set before.

Here's the new default behaviour:

1. First, we check if the input module contains any layers known to be incompatible with the DP-SGD (e.g. BatchNorm). Note, that these restrictions are fundamental to how DP-SGD works and will always be revelant
2. Then, for each layer we select a method of computing per sample gradients. For performance reasons, we still use old manually written grad samplers for the layers we support and fall back to the generic functorch-based grad sampler for all other layers.

You can also force functorch-based grad sampler for every layer by passing grad_sample_mode="functorch" to PrivacyEngine.make_private() or force_functorch=False to GradSampleModule's constructor.

If you're using functorch for your training pipeline already, consider using GradSampleModuleNoOp (grad_sample_mode="no_op") . As suggested by the name, is performs no action and expects client to compute per sample gradients themselves. See our CIFAR-10 example for code demonstration.

Note, that this functionality is still in beta and we haven't fully explored it's limitations. Please report any weird behaviour or inconsistencies you encounter to out github issues, we greatly appreciate the feedback.

ExpandedWeights - yet another way to compute per sample gradients

One more exciting feature now available in core PyTorch is ExpandedWeights. This feature uses old Opacus' approach of manually-written vectorized per sample gradient computations, but achieves much better performance.

To activate ExpandedWeights pass grad_sample_mode="ew" to PrivacyEngine.make_private() or use GradSampleModuleExpandedWeights

Summary: 3 different ways to compute per sample gradients

With the recent updates, Opacus now supports 3 different ways to compute per sample gradients. Below is the quick comparison. For more details refer to the grad sample README.md

TL;DR: If you want stable implementation, use GradSampleModule (grad_sample_mode="hooks").
If you want to experiment with the new functionality, you have two options. Try
GradSampleModuleExpandedWeights(grad_sample_mode="ew") for better performance and grad_sample_mode=functorch
if your model is not supported by GradSampleModule.

Please switch back to GradSampleModule(grad_sample_mode="hooks") if you encounter strange errors or unexpexted behaviour.
We'd also appreciate it if you report these to us

xxx	Hooks	Expanded Weights	Functorch
Required PyTorch version	1.8+	1.13+	1.12 (to be updated)
Development status	Underlying mechanism deprecated	Beta	Beta
Runtime Performance†	baseline	✅ ~25% faster	🟨 0-50% slower
Any DP-allowed†† layers	Not supported	Not supported	✅ Supported
Most popular nn.* layers	✅ Supported	✅ Supported	✅ Supported
torchscripted models	Not supported	✅ Supported	Not supported
Client-provided grad sampler	✅ Supported	Not supported	✅ Not needed
batch_first=False	✅ Supported	Not supported	✅ Supported
Recurrent networks	✅ Supported	Not supported	✅ Supported
Padding same in Conv	✅ Supported	Not supported	✅ Supported

† Note, that performance differences are unstable and can vary a lot depending on the exact model and batch size.
Numbers above are averaged over benchmarks with small models consisting of convolutional and linear layers.
Note, that performance differences are only observed on GPU training, CPU performance seem to be almost identical
for all approaches.

†† Layers that produce joint computations on batch samples (e.g. BatchNorm) are not allowed under any approach

Other improvements

• Fix utils.unfold2d with non-symmetric pad/dilation/kernel_size/stride (#443)
• Add support for "same" and "valid" padding for hooks-based grad sampler for convolution layers
• Improve model validation to support frozen layers and catch copied parameters (#489)
• Remove annoying logging from set_to_none (#471)
• Improved documentation (#480, #478, #482, #485, #486, #487, #488)
• Imtegration test improvements (#407, #479, #481. #473)